IMRN International Mathematics Research Notices1999, No. 5
Weight Modules of Direct Limit Lie Algebras
Ivan Dimitrov and Ivan Penkov
Contents
0 Introduction 223
1 Generalities on triangular decompositions and weight modules 224
2 Direct limits of reductive Lie algebras 226
3 The shadow of an irreducible weight module 231
4 Integrable modules 237
5 Pseudo highest weight modules 242
6 The support of an arbitrary irreducible weight module 244
0 Introduction
The purpose of this paper is to initiate a systematic study of the irreducible weight rep-
resentations of direct limits of reductive Lie algebras and, in particular, of the classical
simple direct limit Lie algebras A(∞), B(∞), C(∞), andD(∞). We study arbitrary, not nec-
essarily highest weight, irreducible weight modules and describe the supports of all such
modules. The representation theory of the classical direct limit groups has been initi-
ated in the pioneering works of G. Olshanskii [O1], [O2] and is now in an active phase (see
the recent works of A. Habib [Ha], L. Natarajan [Na], K.-H. Neeb [Ne], and L. Natarajan,
E. Rodriguez-Carrington, and J. A. Wolf [NRW]). Nevertheless, the structure theory of
weight representations of the simple direct limit Lie algebras has until recently been
still in its infancy as only highest weight modules have been discussed in written works
(see the works of Yu. A. Bahturin and G. Benkart [BB], K.-H. Neeb [Ne], and T. D. Palev [P]).
Our approach is based mainly on the recent paper [DMP] in which a general
method for studying the support of weight representations of finite-dimensional Lie al-
Received 21 October 1998.Communicated by Yu. I. Manin.
224 Dimitrov and Penkov
gebras (and Lie superalgebras) was developed. We prove first that the shadow of any
irreducible weight module M (over any root reductive direct limit Lie algebra, in partic-
ular over A(∞), B(∞), C(∞), and D(∞)) is well defined. This means that for a given root
α, the intersection of the ray λ + R+α with the support of M, suppM, is either finite
for all λ ∈ suppM or infinite for all λ ∈ suppM. Using this remarkable property of the
support, we assign to M a canonical parabolic subalgebra pM of g and then compare M
with the irreducible quotient of a certain g-module induced from pM. In the case of a
finite-dimensional Lie algebra, the Fernando-Futorny parabolic induction theorem (see
[Fe], [Fu], and [DMP]) states thatM is always such a quotient and,moreover, that suppM
is nothing but the support of the induced module. In the direct limit case, we show that
this is no longer true but nevertheless, using the fact that the shadow is well defined,we
present a description of the support of any irreducible weight module.
We discuss in more detail the following special cases: where suppM is finite in
all root directions (these are the finite integrable irreducible modules and they are ana-
logues of finite-dimensional irreducible modules over finite-dimensional Lie algebras);
and the more general case where suppM is finite in at least one of each of two mutually
opposite root directions. An interesting feature in the first case is that M is not nec-
essarily a highest weight module; i.e., the analogues of finite-dimensional irreducible
modules are already outside the class of highest weight modules. We present an explicit
parametrization of all finite integrable irreducible modules. The modules corresponding
to the second case are by definition pseudo highest weight modules. One of their inter-
esting features is that, in general, they are not obtained by parabolic induction as in the
Fernando-Futorny theorem.
1 Generalities on triangular decompositions and
weight modules
The ground field is C. The signs ⊂+ and ⊃+ denote the semidirect sum of Lie algebras
(if g = g′⊂+ g′′, then g′ is an ideal in g as well as a g′′-module). The superscript ∗ always
stands for dual space. We set R+ := {r ∈ R | r ≥ 0}, R− := −R+, Z± := Z ∩ R±, and the
linear span with coefficients in C (respectively, R, Z, R±) is denoted by 〈 〉C (resp., by
〈 〉R, 〈 〉Z, 〈 〉R± ). If g is any Lie algebra and M is a g-module, we call M integrable if
and only if g acts locally finitely on M; i.e., if for any m ∈ M and any g ∈ g, the space
〈m,g·m,g2 ·m, . . .〉C is finite-dimensional. This terminology has been introduced by V. Kac;
in [BB], integrable modules are called locally finite. If g is a direct sum of Lie algebras,
g = ⊕s∈Sgs, and for each s, Ms is an irreducible gs-module with a fixed nonzero vector
ms ∈ M, then (⊗sMs)(⊗sms) denotes the g-submodule of ⊗sMs generated by ⊗sms. It is
easy to check that (⊗sMs)(⊗sms) is an irreducible g-module.
Weight Modules of Direct Limit Lie Algebras 225
Let g be a Lie algebra. A Cartan subalgebra of g is by definition a self-normalizing
nilpotent Lie subalgebra h ⊂ g. We do not assume g or h to be finite-dimensional. A g-
moduleM is a generalized weight g-module if and only if as an h-module, M decomposes
as the direct sum ⊕λ∈h∗Mλ, where
Mλ := {m ∈M | h− λ(h) acts nilpotently on m for every h ∈ h} .
We call Mλ the generalized weight space of M of weight λ. Obviously, a generalized
weight g-module is integrable as an h-module. The support ofM, suppM, consists of all
weights λwithMλ 6= 0. A generalized weight moduleM is a weight module if and only if
h acts semisimply on M, i.e., if and only if each generalized weight space is isomorphic
to the direct sum of one-dimensional h-modules.
Henceforth, g will denote a Lie algebra with a fixed Cartan subalgebra h such that
g is a generalized weight module; i.e.,
g = h⊕ (⊕α∈h∗\{0}gα). (1)
The generalized weight spaces gα are by definition the root spaces of g; ∆ := {α ∈h∗\{0}|gα 6= 0} is the set of roots of g; and the decomposition (1) is the root decompo-
sition of g.
Below we recall the definitions of a triangular decomposition of ∆ and of a Borel
subalgebra for an arbitrary Lie algebra g with a root decomposition. For more details,
see [DP2]. A decomposition of ∆,
∆ = ∆+ t ∆−, (2)
is a triangular decomposition of ∆ if and only if the cone 〈∆+∪−∆−〉R+ 1 (or equivalently,
its opposite cone 〈−∆+ ∪ ∆−〉R+ ) contains no (real) vector subspace. Equivalently, (2) is a
triangular decomposition if the following is a well-defined R-linear partial order on 〈∆〉R:
η ≥ µ⇔ η = µ+n∑i=1
ciαi for some αi ∈ ∆+ ∪ −∆− and ci ∈ R+.
(A partial order isR-linear if it is compatible with addition and multiplication by positive
real numbers, and if multiplication by negative real numbers changes order direction. In
what follows, unless explicitly stated that it is partial, an order will always be assumed
linear, i.e., such that α 6= β implies α < β or α > β. In particular, an R-linear order is
by definition an order which is in addition R-linear.) Any regular real hyperplane H in
1We use the term cone as a synonym for an R+-invariant additive subset of a real vector space.
226 Dimitrov and Penkov
〈∆〉R (i.e., a codimension one linear subspace H with H ∩ ∆ = ∅) determines exactly two
triangular decompositions of ∆: we first assign (in an arbitrary way) the sign + to one
of the two connected components of 〈∆〉R\H, and the sign − to the other. ∆± are then by
definition the subsets of ∆which belong, respectively, to the “positive” and the “negative”
connected components of 〈∆〉R\H. In general, not every triangular decomposition of ∆
arises in this way (see [DP1]). Nevertheless, it is true that every triangular decomposition
is determined by an (not unique) oriented maximal chain of vector subspaces in 〈∆〉R:
see the Appendix, where we establish the precise interrelationship between oriented
maximal chains in 〈∆〉R, R-linear orders on 〈∆〉R, and triangular decompositions of ∆.
A Lie subalgebra b of g is by definition a Borel subalgebra of g if and only if
b = h⊃+ (⊕α∈∆+gα) for some triangular decomposition ∆ = ∆+t∆−. In what follows, a Borel
subalgebra of g always means a Borel subalgebra containing the fixed Cartan subalgebra
h. Adopting terminology from affine Kac-Moody algebras,we call a Borel subalgebra stan-
dard if and only if it corresponds to a triangular decomposition that can be determined
by a regular hyperplane H in 〈∆〉R.
If b is a Borel subalgebra and λ ∈ h∗, then the Verma module Vb(λ) with b-
highest weight λ is by definition the induced module U(g) ⊗U(b) Cλ, Cλ being the one-
dimensional b-module on which h acts via λ. Any quotient of Vb(λ) is by definition a
b-highest weight module. Furthermore, Vb(λ) has a unique maximal proper g-submodule
Ib(λ), and Vb(λ) := Vb(λ)/Ib(λ) is by definition the irreducible b-highest weight g-module
with highest weight λ.
2 Direct limits of reductive Lie algebras
A homomorphism ϕ : g→ g′ of Lie algebras with root decomposition is a root homomor-
phism if and only if ϕ(h) ⊂ h′ (h and h′ being the corresponding fixed Cartan subalgebras)
and ϕ maps any root space of g into a root space of g′. Let
g1ϕ1−→ g2
ϕ2−→ . . .ϕn−1−→ gn
ϕn−→ . . . (3)
be a chain of homomorphisms of finite-dimensional Lie algebras, and let g := lim−→ gn be
the direct limit Lie algebra. We say that g is a root direct limit of the system (3) if and
only if ϕn is a root homomorphism for every n. In the latter case, hn denotes the fixed
Cartan subalgebra of gn. We define a Lie algebra g to be a root direct limit Lie algebra if
and only if g is a root direct limit of some direct system of the form (3). Furthermore, a
nonzero Lie algebra g is a root reductive direct limit of the system (3) if and only if all
gn are reductive; and g is a root simple direct limit Lie algebra if and only if g is a root
direct limit of a system (3) in which all gn are simple.
Weight Modules of Direct Limit Lie Algebras 227
Proposition 1. Let g be a root direct limit Lie algebra. Then h := lim−→ hn is a Cartan
subalgebra of g and g has a root decomposition with respect to h such that ∆ = lim−→∆n,
where ∆n and ∆ are, respectively, the roots of gn and g. If all root homomorphisms ϕn
are embeddings, one has simply h = ∪nhn and ∆ = ∪n∆n.
Proof. The proof is a trivial exercise.
Every simple finite-dimensional Lie algebra g is a root simple direct limit Lie
algebra: we set gn := g and ϕn := idg. To define the simple Lie algebras A(∞), B(∞),
C(∞), and D(∞), it suffices to let gn be the corresponding rank n simple Lie algebra and
to request that all ϕn be injective root homomorphisms, i.e., root embeddings. Indeed,
one can check that in these cases, the direct limit Lie algebra does not depend up to
isomorphism on the choice of root embeddings ϕn. More generally, Theorem 4.4 in [BB]
implies that every infinite-dimensional root simple direct limit Lie algebra is isomorphic
to one of the Lie algebras A(∞), B(∞), C(∞), or D(∞). Note however, that for general
root reductive direct limit Lie algebras, the direct limit Lie algebra can depend on the
choice of root homomorphisms ϕn even if they are embeddings. Indeed, set, for instance,
gn := A(2n)⊕ B(2n), and let ϕn,ϕ′n : gn → gn+1 be root embeddings such that ϕn(A(2n)) ⊂A(2n+1) and ϕn(B(2n)) ⊂ B(2n+1) but ϕ′n(A(2n)⊕ B(2n)) ⊂ B(2n+1). Then g ' A(∞)⊕ B(∞) but
g′ ' B(∞).
Let g be a root reductive direct limit Lie algebra and let k be a Lie subalgebra of
g such that h ⊂ k. Then ∆k ⊂ ∆g and we can set ∆ssk
:= ∆k ∩ (−∆k). The Lie subalgebra
kss := [⊕α∈∆ssk
kα,⊕α∈∆ssk
kα] of k is an analogue of the semisimple part of k in the case where
k is not finite-dimensional.
Theorem 1. Let g be a root reductive direct limit Lie algebra. Then we have the following.
(i) g ' gss⊂+ A, where A is an abelian Lie algebra of finite or countable dimension.
Furthermore, h = hss ⊕A, where hss := h ∩ gss.
(ii) gss ' ⊕s∈Sgs, where S is a finite or countable family of root simple direct limit
Lie algebras gs.
Proof. (i) Decompose h as h = hss ⊕ A for some vector space A. Then A is an abelian
subalgebra of h (since h itself is abelian) of at most countable dimension, and using
Proposition 1, one checks that g ' gss⊂+ A.
(ii) Let g = lim−→ gn. Proposition 1 implies immediately that gss = lim−→ gssn . Furthermore,
since a Cartan subalgebra hn ⊂ gn is fixed for every n, there is a canonical decomposition
gn = (⊕t∈Sngt)⊕ Znsuch that all gt are simple, Zn is abelian, and, for every α ∈ ∆n, gαn ⊂ gt for some t ∈ Sn.
Then, for any t ∈ Sn, either ϕn(gt) = 0 or there is t′ ∈ Sn+1 so that ϕn(gt) is a nontrivial
228 Dimitrov and Penkov
subalgebra of gt′
(we assume that the sets Sn are pairwise disjoint). Put S′ := ∪n{t ∈ Sn |ϕn,n′ (gt) 6= 0 for everyn′ > n}, where ϕn1,n2 := ϕn2 ◦ . . . ◦ ϕn1 for n1 < n2 . Introduce an
equivalence relation ∼ on S′ by setting t1 ∼ t2 for t1 ∈ Sn1 , t2 ∈ Sn2 , if and only if there
exists n such that ϕn1,n(gt1 ) and ϕn2,n(gt2 ) belong to the same simple component of gn.
Define S as the set of classes of ∼-equivalence. For every s ∈ S, the set {gt}t∈s is partially
ordered by the maps ϕn1,n2 : gt1 → gt2 . Let gs be the direct limit Lie algebra of a maximal
chain of Lie algebras among {gt}t∈s with respect to this partial order. Obviously, gs is a
root simple direct limit Lie algebra, and it does not depend on the choice of the maximal
chain. Finally, one checks easily that gss = lim−→ gssn ' ⊕s∈Sgs.
Example 1. gl(∞) can be defined as the Lie algebra of all infinite matrices (aij)i, j∈Z+ with
finitely many nonzero entries. Then gl(∞) ' A(∞)⊂+ C, but gl(∞) 6' A(∞)⊕C as the center
of gl(∞) is trivial.
Theorem 1 gives an almost explicit description of all root reductive direct limit
Lie algebras. In particular, it implies that any root space of a root reductive direct limit
Lie algebra has dimension one. Moreover, if π : h∗ → (hss)∗ denotes the natural projection,
then Theorem 1 implies that π induces a bijection between ∆ and the set of roots of gss.
Note also that for any λ ∈ h∗ and any α ∈ ∆, the scalar product (λ|hn , α|hn )n ((·, ·)n being
the standard bilinear form h∗n × h∗n → C) has a fixed value for all n such that α|hn 6= 0,
and therefore we can set (λ, α) := (λ|hn′ α|hn ) for sufficiently large n. We put then
〈λ, α〉 := 2(λ, α)
(α,α).
By definition, λ is an integral weight of g if and only if 〈λ, α〉 ∈ Z for every α ∈ Z; and λ is
b-dominant, for a Borel subalgebra b = h(⊕α∈∆+gα), if and only if 〈λ, α〉 ≥ 0 for all α ∈ ∆+.
The Weyl group W of g is the group generated by all reflections σα: h∗ → h∗, α ∈ ∆,where
σα := λ− 〈λ, α〉α. By its very definition, W acts on h∗.
The exact relationship between irreducible generalized weight g-modules and
irreducible generalized weight gss-modules (all of which automatically turn out to be
weight modules) is established in the following proposition.
Proposition 2. (i) Every irreducible generalized weight g-moduleM is a weight module.
(ii) Every irreducible weight g-module M is irreducible as a (weight) gss-module.
(iii) Given any irreducible weight gss-moduleMss, every λ ∈ h∗withπ(λ) ∈ suppMss
defines a unique structure of an irreducible weight g-module on Mss which extends the
gss-module structure on Mss. If Mss(λ) denotes the resulting g-module, then Mss(λ) 'Mss(λ′) if and only if π(λ− λ′) =∑i ciπ(αi) for some αi ∈ ∆ implies λ− λ′ =∑i ciαi.
Proof. (i) Let U0 denote the subalgebra of the enveloping algebra U(g) generated by
all monomials of weight zero. Since h acts semisimply on g, the symmetric algebra S(h)
Weight Modules of Direct Limit Lie Algebras 229
belongs to the center of U0. Furthermore, any generalized weight space Mλ is an irre-
ducible U0-module and, by a general version of Schur’s Lemma, S(h) acts via a scalar on
Mλ. Therefore, M is a semisimple h-module; i.e., M is a weight module.
(ii) Follows immediately from (i) and Theorem 1(i).
(iii) Let any a ∈ A act on the weight space Mµ of Mss via multiplication by λ(a)+∑i αi(a), where µ − π(λ) =∑i ciπ(αi). It is straightforward to verify that this equips Mss
with a well-defined g-module structure.
The isomorphism criterion is also an easy exercise.
Our main objective in this paper is the study of the irreducible weight modules
over root reductive direct limit Lie algebras. The case of finite-dimensional reductive Lie
algebras is discussed in particular in [Fe], [Fu], [DMP], [M].
In the rest of this section,we study the structure of the root simple direct limit Lie
algebras, and g stands forA(∞), B(∞), C(∞), orD(∞). If εi are the usual linear functions on
the Cartan subalgebras of the simple finite-dimensional Lie algebras (see for example, [B]
or [Hu]), one can let i run from 1 to∞ and then (using Proposition 1) verify the following
list of roots:
A(∞): ∆ = {εi − εj | i 6= j},B(∞): ∆ = {±εi,±εi ± εj | i 6= j},C(∞): ∆ = {±2εi,±εi ± εj | i 6= j},D(∞): ∆ = {±εi ± εj | i 6= j}.
Every sequence of complex numbers {λn}n=1,2,... determines a weight λ of g by setting
λ(εn) := λn. For g = B(∞), C(∞), or D(∞), every weight λ of g recovers the sequence
{λn := λ(εn)}n=1,2,...; for g = A(∞), the weight λ recovers the sequence {λn := λ(εn)}n=1,2,... up
to an additive constant only.
It is proved in [DP2, Proposition 2] that all Borel subalgebras ofA(∞), B(∞), C(∞),
andD(∞) are standard. More precisely, for every Borel subalgebra b of g, there is a linear
function ϕ : 〈∆〉R→ R such that b = h⊕ (⊕ϕ(α)>0gα). Define an order (or a partial order) on
the set {0} ∪ {±εi} to be Z2-linear if and only if multiplication by −1 reverses the order.
Then [DP2, Proposition 2] is essentially equivalent to the following statement.
Proposition 3. If g = A(∞), then there is a bijection between Borel subalgebras of g and
orders on the set {εi}.If g = B(∞) or C(∞), there is a bijection between Borel subalgebras of g and
Z2-linear orders on the set {0} ∪ {±εi}.
230 Dimitrov and Penkov
If g = D(∞), there is a bijection between Borel subalgebras of g and Z2-linear
orders on the set {0} ∪ {±εi}with the property that, if there is a minimal positive element
with respect to this order, then this element is of the form εi.
Proof. The pullback via ϕ of the standard order on R determines an order on {0} ∪ ∆which induces an order, respectively, on {εi} or {0}∪{±εi}, as desired. Conversely, for every
order {εi}, or, respectively, for every Z2-linear order on {0} ∪ {±εi} as in the proposition,
there exists a (nonunique) linear function ϕ : 〈{εi}〉R→ R such that ∆+ = ϕ−1(R+)∩∆.
The result of Proposition 3 can be found in an equivalent form in [Ne]. The case
of A(∞) is due to V. Kac (unpublished).
Example 2. A(∞) is naturally identified with the Lie algebra of traceless infinite ma-
trices (ai, j)i, j∈Z+ with finitely many nonzero entries, as well as with the Lie algebra of
traceless double infinite matrices (ai, j)i, j∈Z with finitely many nonzero entries. The re-
spective algebras of upper triangular matrices are nonisomorphic Borel subalgebras of
A(∞) corresponding, respectively, to the orders ε1 > ε2 > ε3 > · · · and · · · > ε6 > ε4 > ε2 >
ε1 > ε3 > ε5 > · · ·.A parabolic subalgebra of a root reductive direct limit Lie algebra is by definition
a Lie subalgebra containing a Borel subalgebra. (The general definition of a parabolic
subalgebra of a Lie algebra with root decomposition is given in the Appendix.) Here is
an explicit description of all parabolic subalgebras of A(∞), B(∞), C(∞), and D(∞).
Proposition 4. If g = A(∞), then there is a bijection between parabolic subalgebras of
g and partial orders on the set {εi}.If g = B(∞) or C(∞), then there is a bijection between parabolic subalgebras of g
and Z2-linear partial orders on the set {0} ∪ {±εi}.If g = D(∞), then there is a bijection between parabolic subalgebras of g and Z2-
linear partial orders on the set {0} ∪ {±εi} with the property that if εi is not comparable
with 0 for some i (i.e., neither εi > 0 nor εi < 0), then εj is also not comparable with 0 for
some j 6= i.
Proof. Given a partial order on {εi} (respectively, a Z2-linear partial order on {0} ∪ {±εi}with the additional property for g = D(∞)), it determines a unique partial order > on the
set {0} ∪ ∆. Put p> := h ⊕ (⊕α>0 or α not comparable with 0g
α). Then p> is the parabolic
subalgebra corresponding to the initial partial order. Conversely, let p be a parabolic
subalgebra. For α ∈ ∆, set α >p 0 if and only if gα ⊂ p but g−α 6⊂ p. Using the explicit form
of ∆, it is easy to verify that this determines a unique partial order on {εi} (respectively,
on {0} ∪ {±εi}, as desired).
Weight Modules of Direct Limit Lie Algebras 231
The next proposition describes pss for any parabolic subalgebra p of g = A(∞),
B(∞), C(∞), or D(∞); it will be used in Section 3.
Proposition 5. Let p be a parabolic subalgebra of g.
(i) pss is isomorphic to a direct sum of simple Lie algebras, each of which is one
of the following:
• A(n) or A(∞), if g = A(∞);
• A(n), A(∞), B(n), or B(∞) with at most one simple component isomorphic to
B(n) or B(∞), if g = B(∞);
• A(n), A(∞), C(n), or C(∞) with at most one simple component isomorphic to
C(n) or C(∞), if g = C(∞);
• A(n), A(∞), D(n), or D(∞) with at most one simple component isomorphic to
D(n) or D(∞), if g = D(∞).
(ii) If pss 6' 0, then pss+ h ' (⊕t∈T kt)⊕Z,where Z is abelian and kt is isomorphic to
gl(n) or gl(∞) for any t ∈ T except at most one index t0 ∈ T for which kt0 is a root simple
direct limit Lie algebra.
Proof. (i) Let >p, as in Proposition 4, be the partial order corresponding to p. Partition
the set {εi} (respectively, {0} ∪ {±εi}) into subsets in such a way that two elements are
comparable with respect to >p if and only if they belong to different subsets. Denote by
S′ the resulting set of subsets of {εi} (respectively, of {0} ∪ {±εi}). Let furthermore S be
a subset of S′ which contains exactly one element of any pair s, s′ of mutually opposite
elements of S′; and for every s ∈ S, define gs to be the Lie algebra generated by gα, where
α 6= 0 belongs to s or is a sum of any two elements of s. It is straightforward to verify
that pss ' ⊕s∈S,gs 6=0gs is the decomposition of pss into a direct sum of root simple direct
limit Lie algebras and that this decomposition satisfies (i).
(ii) The main point is to notice that if s 6= −s, then gs ' A(n) for somen or gs ' A(∞)
and, furthermore, that there is at most one s ∈ S such that s = −s. To complete the proof,
it remains to show that each of the Lie subalgebras gs for s 6= −s can be extended to a
Lie subalgebra isomorphic to gl(n) or gl(∞). This latter argument is purely combinatorial
and we leave it to the reader.
3 The shadow of an irreducible weight module
Let g be a Lie algebra with a root decomposition, let M be an irreducible generalized
weight g-module, and let λ be a fixed point in suppM. For any α ∈ ∆, consider the set
mλα := {q ∈ R | λ + qα ∈ suppM} ⊂ R. There are four possible types of sets mλ
α, as
follows: bounded in both directions; unbounded in both directions; bounded from above
232 Dimitrov and Penkov
but unbounded from below; unbounded from above but bounded from below. It is proved
in [DMP] that, if g is finite-dimensional, then the type of mλα depends only on α and not
on λ, and therefore the module M itself determines a partition of ∆ into four mutually
disjoint subsets:
∆FM := {α ∈ ∆ | mλα is bounded in both directions
},
∆IM := {α ∈ ∆ | mλα is unbounded in both directions
},
∆+M := {α ∈ ∆ | mλα is bounded from above and unbounded from below
},
∆−M := {α ∈ ∆ | mλα is bounded from below and unbounded from above
}.
The corresponding decomposition
g = (gFM + gIM
)⊕ g+M ⊕ g
−M, (4)
where gFM := h⊕(⊕α∈∆FM
gα), gIM := h⊕(⊕α∈∆IM
gα),and g±M := ⊕α∈∆±
Mgα, is theM-decomposition
of g. The triple (gIM, g+M, g
−M) is the shadow ofM onto g. If g is infinite-dimensional,we say
that the shadow ofM onto g is well defined if it is true that the type ofmλα depends only
on α and not on λ. Then the decomposition (4) (as well as the triple (gIM, g+M, g
−M)) is well
defined.
In the case where g is finite-dimensional and reductive, it is furthermore true
that pM := (gFM+ gIM)⊕ g+M is a parabolic subalgebra of g whose reductive part is gFM+ gIM,
and that there is a natural surjection
ϕM : U(g)⊗U(pM) Mg+M →M,
where Mg+M is the irreducible (gFM + gIM)-submodule of M which consists of all vectors
in M annihilated by g+M. Moreover, suppM simply coincides with supp(U(g)⊗U(pM) M
g+M ).
This is the Fernando-Futorny parabolic induction theorem (see [Fe] and also [DMP]), and
in particular it provides an explicit description of suppM.
The main purpose of this paper is to understand analogues of these results for
a root reductive direct limit Lie algebra g. Roughly, the situation turns out to be as fol-
lows: the shadow of an arbitrary irreducible g-module M exists and defines a parabolic
subalgebra pM of g; however, the parabolic induction theorem does not hold. The rea-
son for the latter is that, as the nilpotent Lie algebra g+M is infinite-dimensional, it may
not annihilate any nonzero vector in M; such a module M is constructed in Example 5.
Nevertheless, the existence of the shadow and the direct limit structure on g enable us
to obtain an explicit description of suppM for any M.
Weight Modules of Direct Limit Lie Algebras 233
Throughout the rest of this paper (except in the Appendix), g := lim−→ gn will be a
root reductive direct limit Lie algebra such that all ϕn are embeddings, and M will be a
fixed irreducible weight module over g.
Theorem 2. The shadow of M is well defined.
Proof. Note first that, for any given λ ∈ suppM, there exist irreducible (gn+ h)-modules
Mn such that λ ∈ suppMn for every n, and suppM = ∪n suppMn. Indeed, fix m ∈ Mλ,
m 6= 0 and define Mn as any irreducible quotient of the (gn + h)-module U(gn + h) ·m. One
checks immediately that suppM = ∪n suppMn.
Fixα ∈ ∆. To prove the theorem,we need to show that ifmλα is bounded from above
for some λ ∈ suppM (the proof for the case whenmλα is bounded from below is exactly the
same), then mµα is bounded from above for any other µ ∈ suppM. First, let λ′ = λ+ kα be
the end point of the α-string through λ in suppM. Then fix µ ∈ suppM and pickN so that
µ ∈ suppMN, λ′ ∈ suppMN, and α ∈ ∆N. Consider now the α-string through µ. It is the
union of the α-strings through µ in suppMn for n = N,N+1, . . .. The parabolic induction
theorem implies that suppMn (and, in particular, the α-string through µ in suppMn) is
contained entirely in an affine half-space in λ+〈∆n〉R whose boundary (affine) hyperplane
Hn contains λ′ and is spanned by vectors in ∆n. Furthermore, H′n := Hn ∩ (λ + 〈∆N〉R) is
a hyperplane in λ + 〈∆N〉R which contains λ′ and is spanned by vectors in ∆N. Since the
α-string through µ belongs to λ + 〈∆N〉R, the α-string through µ in suppMn is bounded
by H′n. But there are only finitely many hyperplanes in λ+ 〈∆N〉R passing through λ′ and
spanned by vectors in ∆N and hence among the hyperplanes H′n for n = N,N+ 1, . . . only
finitely many are different. Therefore, the α-strings through µ in suppMn are uniformly
bounded from above; i.e., mµα is bounded from above.
The next theorem describes the structure of theM-decomposition and is an exact
analogue of the corresponding theorem for finite-dimensional reductive Lie algebras.
Theorem 3. (i) gFM, gIM, g
+M, and g
−M are Lie subalgebras of g, and h is a Cartan subalgebra
for both gFM and gIM.
(ii) [(gFM)ss, (gIM)ss] = 0 and therefore gFIM := gFM + gIM is a Lie subalgebra of g.
(iii) g+M and g
−M are gFIM-modules.
(iv) pM := gFIM ⊕ g+M and gFIM ⊕ g
−M are (mutually opposite) parabolic subalgebras
of g.
Proof. (i) Let α,β ∈ ∆ be such that α+β ∈ ∆. Lemma 2 in [PS] implies that ifmλα andmλ
β
are bounded from above, then so ismλα+β. Furthermore, noting that suppM = ∪n suppMn
for some irreducible (gn+h)-modulesMn (see the proof of Theorem 2) and that the support
of every irreducible weight (gn+h)-module is convex,we conclude that suppM is convex.
234 Dimitrov and Penkov
Therefore if mλα and mλ
β are unbounded from above, so is mλα+β. These two facts imply
immediately that all four subspaces gFM, gIM, g+M, and g
−M are subalgebras of g. The fact
that h is a Cartan subalgebra for both gFM and gIM is obvious.
(ii) If α′ ∈ ∆FM and β′ ∈ ∆IM, then α′ + β′ 6∈ ∆. Indeed, if α′ + β′ ∈ ∆ and mλα′+β′ is
bounded from above, thenmλβ′ would be bounded from above because β′ = −α′ + (α′ +β′).
If, on the other hand, α′ +β′ ∈ ∆ andmλα′+β′ is unbounded from above, thenmλ
α′ would be
unbounded from above because α′ = −β′ + (α′ + β′). Since both of these conclusions
contradict the choice of α′ and β′, we obtain that α′ + β′ 6∈ ∆ and thus that [(gFM)ss,
(gIM)ss] = 0.
(iii) If α′ ∈ ∆FM, β′ ∈ ∆+M, and α′ +β′ ∈ ∆, then againmλα′+β′ is bounded from above.
Assuming that mλα′+β′ is bounded from below, we obtain that mλ
β′ is bounded from below
as well since−β′ = −(α′ +β′)+α′,which contradicts the fact that β′ ∈ ∆+M. Hencemλα′+β′ is
unbounded from below and α′ +β′ ∈ ∆+M. This proves that g+M is a gFM-module. One shows
in a similar way that g±M are gFM- and gIM-modules.
(iv) is a direct corollary of (i), (ii), and (iii).
We define an irreducible weight g-module M to be cuspidal if and only if g = gIM.
In the rest of this section, we prove that for every parabolic subalgebra p of g,
there is an irreducible g-module M such that p = pM. Indeed, there is the following more
general theorem.
Theorem 4. Let g be a root reductive direct limit Lie algebra. For any given splitting
∆ = ∆Ft∆It∆+t∆− with∆F = −∆F, ∆I = −∆I, ∆− = −∆+, and such that its corresponding
decomposition
g = (gF + gI)⊕ g+ ⊕ g− (5)
satisfies properties (i)–(iii) of Theorem 3, there exists an irreducible weight moduleM for
which (5) is the M-decomposition of g, i.e., for which gFM = gF, gIM = gI, and g±M = g±.
Proof. We start with the observation that it suffices to prove the theorem for a root
simple direct limit Lie algebra. Indeed, let g ' (⊕s∈Sgs)⊂+ A be as in Theorem 1. Assume,
furthermore, that for each s, Ms is an irreducible weight gs-module corresponding (as in
the theorem) to the restriction of decomposition (5) to gs. Fix nonzero vectors ms ∈ Ms.
Then the reader will verify straightforwardly that, for any pair (Mss := (⊗sMs)(⊗sms), λ)
as in Proposition 2(iii), the g-moduleM :=Mss(λ) is as required by the theorem. Therefore,
in the rest of the proof, we will assume that g is a root simple direct limit Lie algebra.
We will prove first that cuspidal modules exist.
Weight Modules of Direct Limit Lie Algebras 235
Lemma 1. Let g be a simple finite-dimensional Lie algebra. For any given weight µ ∈ h∗,
there exists a cuspidal g-moduleM such that the center Z ofU(g) acts onM via the central
character χµ : Z → C (obtained by extending µ to a homomorphism µ : S·(h) → C and
composing with Harish-Chandra’s homomorphism Z→ S·(h)).
Proof of Lemma 1. It goes by induction on the rank of g. It is a classical fact that sl(2)
admits a cuspidal module of any given central character, so it remains to make the
induction step. Let g′ be a reductive subalgebra of g which contains h and such that the
rank of its semisimple part equals rk g−1. Note that, applying the induction assumption
to the semisimple part of g′ (which clearly is a simple Lie algebra), and then tensoring
with an appropriate one-dimensional representation of the center of g′, we see that the
claim of the lemma is also true for g′. Thus we can fixM′ to be a cuspidal g′-module with
central character χ′µ′ ,where µ′ ∈ h∗ andw′(µ′)−µ 6∈ 〈∆〉Z for anyw′ in the Weyl groupW′ of
g′. Denote byU′ the subalgebra ofU(g) generated byU(g′) and Z. AsU(g) is a free Z-module
(Kostant’s theorem) and U(g′)∩Z = C, U′ is isomorphic to U(g′)⊗CZ. Therefore, the tensor
product M′χµ :=M′ ⊗ Cχµ, Cχµ being the one-dimensional Z-module corresponding to χµ,
is a well-defined U′-module. Consider now any irreducible quotient M of the induced
g-moduleU(g)⊗U′M′χµ . Obviously, M is a weight module and we claim thatM is cuspidal.
Assume the contrary. Since M′ is a cuspidal g′-module and the rank of the semisimple
part of g′ is rk g − 1, the fact that M is not cuspidal means that M′ is a quotient of
U(g)⊗U(p) M′′, where p ⊃ g′ and M′′ is a cuspidal g′-module of central character χ′µ. Then,
sinceU(g) is an integrableU(g′)-module andM′ is a subquotient ofU(g)⊗U(p)M′′, the central
character of M′ equals χw′(µ′)−η for some η ∈ 〈∆〉Z and some w′ ∈W′. But as χw′(µ′)−η = χ′µ′ ,we have w′(µ′) − η ∈ W′ · µ′; i.e., w′(µ′) − µ ∈ 〈∆〉Z. This contradiction implies that M is
cuspidal.
If g = gI, g = lim−→ gn being an infinite-dimensional root simple direct limit Lie
algebra, we can assume that rk gn+1 − rk gn = 1. We then construct MI as the direct limit
lim−→Mn, each Mn being a cuspidal gn-module built by induction precisely as in the proof
of Lemma 1. This proves the theorem in the cuspidal case.
Let now g 6= gI. According to Theorem 1, gF ' ⊕s∈Sgs⊂+ AF and gI ' ⊕t∈Tgt⊂+ AI,where each of the algebras gr for r ∈ S ∪ T is a root simple direct limit Lie algebra and
AF and AI are, respectively, abelian subalgebras of gF and gI. Since the theorem is proved
for the cuspidal case, we can choose a cuspidal irreducible gt module Mt for each t ∈ Tand fix a nonzero vector mt ∈Mt. Then MI := (⊗t∈TMt)(⊗tmt) is a cuspidal (gI)ss-module.
Let M′ denote MI considered as a (gF + gI)ss-module with trivial action of (gF)ss.
First, consider the case where g is finite-dimensional. Denote by WI the Weyl
group of (gI)ss. By Lemma 1, we can assume that the (gI)ss-module MI is chosen in such
236 Dimitrov and Penkov
a way that the central characater of MI corresponds to an orbit WI · η for which the set
WI ·η−η does not contain integral weights of (gI)ss. LetMFI denote MI with its obvious (gF+gI)⊕ g+-module structure, and let M be the irreducible quotient of U(g)⊗U((gF+gI)⊕g+) M
FI.
Then the condition on the central characater of MI ensures that the M-decomposition
of g is nothing but (5). The last statement can be easily verified, which we leave to the
reader. Theorem 4 is therefore proved for a finite-dimensional reductive g.
A direct generalization of this argument does not go through for an infinite-
dimensional g. Instead, there is the following lemma which allows us to avoid referring
to central characters.
Lemma 2. If g is infinite-dimensional, then the (gF + gI)ss-module structure on M′ can
be extended to a (gF + gI)-module structure in such a way that 〈µ′, α〉 /∈ Z for some (and
hence any) µ′ ∈ suppM′ and any α ∈ ∆+ t ∆−.
Proof of Lemma 2. Theorem 3(iv) implies that gFM + gIM = (pM)ss + h. We will present the
proof in the case where gFM+gIM 6= h. The case where gFM+gIM = h is dealt with in a similar
way.
Using Proposition 5(ii), we conclude that gFM + gIM 6= h implies
gFM + gIM ' (⊕r∈Rgr)⊕ g0 ⊕ Z, (6)
where each gr is isomorphic to gl(n) for some n or to gl(∞); R is a finite or countable set
which does not contain 0 (and may be empty); g0 is a root simple direct limit Lie algebra;
and Z is abelian.
Let ∆s be the root system of gs for s ∈ R ∪ {0}. Set sZ := {k ∈ (Z+\{0}) | there is l ∈(Z+\{0}) with εk − εl ∈ ∆s} and #Z := (Z+\{0})\(∪s∈R∪{0} sZ). As explained in Section 1, any
sequence {λn}n=1,2,... ⊂ C determines a weight λ of (gF+ gI)ss; however, λ reconstructs only
the subsequence {λn}n∈∪s∈R∪{0} sZ up to an additive constant for every index from R. Fix now
µ ∈ suppM′ and let {µn}n=1,2,... be a sequence which determines µ. To prove the lemma, it
suffices to find constants ci ∈ C for i ∈ Z with the following properties:
• ck = cl whenever k and l belong to one and the same set sZ;
• ck = 0 for k ∈ 0Z;
• µk+ ck /∈ Z for every k ∈ (Z+\{0}) for which εk ∈ ∆+ t∆−; 2(µk+ ck) /∈ Z for every
k ∈ (Z+\{0}) for which 2εk ∈ ∆+ t∆−; and ±(µk+ ck)± (µl+ cl) /∈ Z for every
k, l ∈ (Z+\{0}) for which ±εk ± εl ∈ ∆+ t ∆−.
Then µ′ will be the weight of (gF + gI) determined by the sequence {µn + cn}n=1,2,....
Consider the set of sequences {ck}k∈P⊂(Z+\{0}) with the three properties as above.
Introduce an order < on this set by putting {c′k′ }k′∈P′ < {c′′k′′ }k′′∈P′′ if and only if P′ ⊂ P′′
Weight Modules of Direct Limit Lie Algebras 237
and c′k = c′′k for every k ∈ P′. The reader will check that every chain (with respect to <)
of sequences is bounded and that for every {ck}k∈P with P 6= (Z+\{0}), there is a sequence
greater than {ck}k∈P. Therefore any maximal element, which exists by Zorn’s lemma, is a
sequence {ck} with the required properties. Lemma 2 is proved.
Assuming that g is infinite-dimensional, let now MFI be M′ considered as a (gF +gI) ⊕ g+-module with trivial action of g+. Then we define M as the (unique) irreducible
quotient of the g-module U(g)⊗U((gF+gI)⊕g+)MFI. Using Lemma 2, the reader will verify that
the M-decomposition of g is precisely the decomposition (5). The proof of Theorem 4 is
therefore complete.
4 Integrable modules
Proposition 6. Let M be integrable. Then we have the following.
(i) g = gFM + gIM; furthermore, g 6= gFM implies dimMλ = ∞ for any λ ∈ suppM.
(ii) If g is infinite-dimensional and simple, M is either cuspidal or g = gFM, and
both cases are possible.
Proof. (i) As a consequence of the integrability ofM, suppM isW-invariant. Thus ∆±M =∅ and g = gFM + gIM. Furthermore, for any α ∈ ∆, the subalgebra gRα of g generated by gα
and g−α is isomorphic to sl(2), and the integrability of M implies that as a gRα- module,
M is isomorphic to a direct sum of finite-dimensional modules. The assumption that
dimMλ < ∞ for some λ ∈ suppM would lead us to the conclusion that λ belongs to
the support of only finitely many of the finite-dimensional gRα-modules, which would
mean that suppM is finite in the direction of both α and −α. Therefore g 6= gFM implies
dimMλ = ∞ for any λ ∈ suppM.
(ii) If both gFM 6= h and gIM 6= h, Theorems 1 and 3 imply that (gFM)ss and (gIM)ss are
proper ideals in g,which is a contradiction. Thus gFM = h or gIM = h; i.e., g = gIM or g = gFM.
Clearly, g = gFM if M = g is the adjoint representation, so the case g = gFM
is obviously possible. To prove that integrable cuspidal modules exist, it is enough
to construct a tower of embeddings of irreducible finite-dimensional gn-modules Mn,
. . . → Mn → Mn+1 → . . ., such that the support of Mn is shorter than the support of
Mn+1 in all root directions of gn. Obviously, M := lim−→Mn is then an integrable cuspidal
g-module. Here is an explicit example for g = A(∞). Let λn = nε1 − nε2 and let bn be
the Borel subalgebra of gn corresponding to the order ε2 < ε3 < . . . < εn+1 < ε1. Set
Mn := Vbn (λn) for n ≥ 2. There is a unique (up to a scalar multiple) embedding of gn-
modules Mn → Mn+1 (this can be verified directly or by using the classical branching
rule (see, for instance, Ch.4, §6 in [Zh])); and furthermore, the support of Mn+1 is seen
238 Dimitrov and Penkov
immediately to be longer than the support of Mn in each root direction of gn. The proof
of Proposition 6 is therefore complete.
Obviously, ifM is integrable and is a highest weight module with respect to some
Borel subalgebra b, then g = gFM. The existence of cuspidal integrable modules is a new
phenomenon. It is different from the case of a finite-dimensional Lie algebra, where any
integrable irreducible module is finite-dimensional and is thus a highest weight module
with respect to every Borel subalgebra. In contrast, for g = A(∞), B(∞), C(∞), and D(∞),
the trivial g-module is the only irreducible g-module that is a highest weight module for
all Borel subalgebras. Moreover, as we will see in Example 3 below, the equality g = gFM
does not guarantee the existence of a Borel subalgebra with respect to which M is a
highest weight module. We defineM to be finite integrable if and only if g = gFM. The rest
of this section is devoted to the study of finite integrable irreducible g-modules M over
an arbitrary root reductive direct limit Lie algebra g.
The simplest type of finite integrable modules are highest weight integrable
modules, and they are studied in the recent papers [BB], [NRW], and [Ne]. In [BB], the
highest weights of integrable highest weight modules are computed explicitly. In [NRW],
the integrable modules Vb(λ) appear in the context of Borel-Weil-Bott’s theorem and are
realized as the unique nonzero cohomology groups of line bundles on G/B. In [Ne], irre-
ducible highest weight modules with respect to general Borel subalgebras are considered;
furthermore, it is proved that their integrability is equivalent to unitarizability. In The-
orem 5 below, we discuss highest weight modules and, in particular, establish a direct
limit version of H. Weyl’s character formula for integrable highest weight modules.
First we need to recall the notion of basis of a Borel subalgebra of g. A subset
Σ ⊂ ∆+ is a basis of b = h ⊕ (⊕α∈∆+gα) if and only if Σ is a linearly independent set and
every element of ∆+ is a linear combination of elements of Σ with nonnegative integer
coefficients; the elements of Σ are then the simple roots of b. Not every Borel subalgebra
admits a basis. For root simple direct limit Lie algebras, a basis of b is the same as a weak
basis in the terminology of [DP2]; and in [DP2], all Borel subalgebras admitting a weak
basis are described. The result of [DP2] implies that b admits a basis if and only if the
corresponding order on {εi}, or, respectively, on {0} ∪ {±εi}, has the following property:
for every pair of elements of {εi} (respectively, of {0} ∪ {±εi}), there are only finitely many
elements between them. This latter criterion has been established also by K.-H. Neeb in
[Ne].
Theorem 5. Let M = Vb(λ).
(i)M is integrable if and only if λ is an integral b-dominant weight. Furthermore,
if M is integrable, then suppM = Cλ, where Cλ is the intersection of the convex hull of
Weight Modules of Direct Limit Lie Algebras 239
W · λ with λ+ 〈∆〉Z.
(ii) M ' Vb′ (λ′) for given b′ and λ′ if and only if there exists w ∈ W for which
λ′ = w(λ) and there is a parabolic subalgebra p of g containing both w(b) and b′ such that
the p-submodule of M generated by Mλ′ is one-dimensional.
(iii) dimMµ < ∞ for all µ ∈ suppM if and only if M ' Vb(λ) for some Borel
subalgebra b of g which admits a basis.
(iv) If M is integrable, b admits a basis, and chM :=∑µ∈suppM dimMµ · eµ is the
formal character of M, we have
D · chM =∑w∈W
(sgnw)ew(λ+ρb)−ρb, (7)
where D = Πα∈∆+ (1− e−α) and ρb ∈ h∗ is a weight for which ρb(α) = 1 for all simple roots
of b.
Proof. (i) The proof is an exercise. In [BB], a more general criterion for the integrability
of Vb(λ) is established for simple direct limit Lie algebras which are not necessarily root
simple direct limit Lie algebras.
(ii) Define pλ ⊃ b as the maximal parabolic subalgebra of g such that the pλ-
submodule of M generated by Mλ is one-dimensional. Then M ' Vb′′ (λ) if and only if b′′
is a subalgebra of pλ. Finally, it is an exercise to check that M ' Vb′ (λ′) if and only if
λ′ = w(λ), and b′ is a subalgebra of pλ′ = w(pλ) for some w ∈W.
(iii) If b admits a basis, then, using the Poincare-Birkhoff-Witt theorem, one ver-
ifies that all weight spaces of the Verma module Vb(λ) are finite-dimensional (for any
λ ∈ h∗), and hence all weight spaces of Vb(λ) are finite-dimensional as well.
Conversely, let M = Vb(λ) be a highest weight module with finite-dimensional
weight spaces. We need to prove the existence of b which admits a basis and such that
M ' Vb(λ). Let pλ be as above and let g ' (⊕s∈Sgs)⊂+ A as in Theorem 1. Clearly, there
is a Borel subalgebra b ⊂ pλ that admits a basis if and only if for every s ∈ S, there is
a Borel subalgebra bs of gs admitting a basis such that bs ⊂ psλ := pλ ∩ gs. Furthermore,
Ms := Vbs (λ|hs ) (where bs := b∩gs and hs := h∩gs) is an irreducible gs-submodule ofM. Thus
it suffices to prove (iii) for the root simple direct limit Lie algebras gs and their highest
weight modulesMs. For a finite-dimensional gs, (iii) is trivial, so we need to consider only
the case where gs = A(∞), B(∞), C(∞), D(∞).
Let >psλ
be the partial order corresponding to psλ and let α ∈ (∆s)+ (∆s being the
root system of gs) be a difference of two elements δ1 and δ2 of {εi} (respectively, of {0} ∪{±εi}). Then it is not difficult to check that, if there are infinitely many elements of {εi}(respectively, of {0} ∪ {±εi}) between δ1 and δ2 with respect to >ps
λ, the weight space of
Ms with weight λ − α is infinite-dimensional. Therefore, the assumption that all Borel
240 Dimitrov and Penkov
subalgebras bs, such that Ms ' Vbs
(λ|hs ) admit no basis, is contradictory; i.e., a Borel
subalgebra bs exists, as required.
(iv) To prove formula (7), it suffices to notice that if b admits a basis, then g can
be represented as g = lim−→ gn, where each of the simple roots of bn = b ∩ gn is a simple
root of bn+1 = b∩ gn+1. Then M = lim−→Vbn (λ|hn ) and, in both sides of (7), terms of the form
c ·eλ+µ for µ ∈ 〈∆n〉R appear only as they appear in respective sides of (7) for the gn-module
Vb∩gn (λ|hn ). Therefore, Weyl’s original formula implies the infinite version (7).
We now turn our attention to general finite integrable modules M. Here is an
example of a finite integrable M which is not a highest weight module with respect to
any Borel subalgebra of g.
Example 3. Let g = A(∞) and let b ⊂ g be the Borel subalgebra of g corresponding to the
order ε1 > ε2 > ε3 > . . .. Set λn := ε1+ · · · + εn−nεn+1. Since λn ∈ suppVbn+1 (λn+1) and the
weight space Vbn+1 (λn+1)λn is one-dimensional, there is a unique (up to a multiplicative
constant) embedding of weight gn-modules Vbn (λn) → Vbn+1 (λn+1). Set M := lim−→Vbn (λn).
Then g = gFM and all weight spaces of M are one-dimensional. M is not a highest weight
module with respect to any Borel subalgebra of g, as no weight of M is a highest weight
of Vbn (λn) for two consecutive n.
The following two theorems provide an explicit parameterization of all finite
integrable modules as well as an explicit description of their supports. Let Wn ⊂ W
denote the Weyl group of gn.
Theorem 6. Let M be finite integrable.
(i) M ' lim−→Mn for some direct system of finite-dimensional irreducible gn-
modules Mn.
(ii) suppM determines M up to isomorphism.
(iii) Fix a Borel subalgebra b of g. Then M ' lim−→Vbn (λn|hn ) for some sequence {λn}of integral weights of g such that λn|hn is a bn-dominant weight of gn and λn belongs to
an edge of the convex hull of Wn+1 · λn+1.
To prove the theorem, we need a lemma which is a weak version of the classical
branching rule and for which we did not find a convenient quotation.
Lemma 3. Let k ⊂ l be finite-dimensional reductive Lie algebras whose Cartan subal-
gebras coincide. Let furthermore N be an irreducible finite-dimensional l-module with a
decomposition N = ⊕iNi into irreducible k-modules. Fix ν ∈ suppN. Then the following
hold.
(i) There is at least one j0 such that ν ∈ suppNj0 , and suppNj0 contains suppNj
for all j with ν ∈ suppNj;
Weight Modules of Direct Limit Lie Algebras 241
(ii) if rk l′ − rk k′ = 1, l′ and k′ being the semisimple parts of l and k, respectively,
then dimNν′ = 1 for any highest weight ν′ of Nj0 .
Proof. (i) It is enough to prove (i) in the case where rk l′ − rk k′ = 1, and then apply
a simple induction argument. Let rk l′ − rk k′ = 1. Denote by Γ1, . . . , Γp the Wk-orbits of
a highest weight of N (Wk being the Weyl group of k). We can assume that Γ1, . . . , Γp
are ordered in a way that the convex hull Gi of Γi−1 ∪ Γi does not intersect with Γj for
j 6= i − 1, i. Gi is a polyhedron with vertices Γi−1 ∪ Γi, and moreover the convex hull of
suppN equals the union ∪pi=2Gi. The edges ofGi are of three types: the edges of the convex
hull of Γi−1; the edges of the convex hull of Γi; and the line segments connecting the bk-
highest weight among Γi−1 with the bk-highest weight among Γi, where bk is any Borel
subalgebra of k. If ν′ is an integral weight which belongs to one of the latter edges, then
suppVbk(ν′) = (ν′ + 〈∆k〉Z) ∩ suppN. This completes the proof of (i).
(ii) Since ν′ belongs to an edge of the convex hull of suppN, we have
dimNν′ = 1.
Proof of Theorem 6. (i) Fix λ ∈ suppM. We claim that there is a unique (up to isomor-
phism) irreducible finite-dimensional gn+h-moduleMλn such that suppMλ
n = (λ+〈∆n〉Z)∩suppM. To check this, it is enough to verify that ((λ+〈∆n〉Z)∩suppM)|hn equals the support
of some irreducible finite-dimensional gn-module. LetM′n be an irreducible gn+h-module
such that λ ∈ suppM′n and ∪n suppM′n = suppM (such modules are constructed in the
proof of Theorem 2). The equality g = gFM implies that (λ+ 〈∆n〉Z) ∩ suppM is a finite set,
and hence there is n′ such that (λ+ 〈∆n〉Z) ∩ suppM is contained in suppM′n′ . Moreover,
(λ+ 〈∆n〉Z)∩ suppM′n′ = (λ+ 〈∆n〉Z)∩ suppM. Applying Lemma 3(i) to the pair of Lie alge-
bras gn + hn′ ⊂ gn′ and the gn′-module M′n′ , we define Mλ
n as the gn + h-module obtained
by extending a gn+hn′-moduleNj0 to a gn+h-module for which λ ∈ suppMλn. Lemma 3(ii)
implies that there is a unique (up to a scalar multiple) embedding Mλn → Mλ
n+1, and
hence the g-module M′ := lim−→Mλn is well defined. Noting that Mλ
n can be embedded as a
submodule of the gn + h-module M, we conclude that M 'M′.(ii) The crucial point is that (according to its construction) Mλ
n depends only on
suppM and λ ∈ suppM. Furthermore, for any λ′ ∈ suppM, there is m such that λ′ ∈suppMλ
m and hence Mλn ' Mλ′
n for n > m. Letting now n go to ∞, and noting that for
any pair λ, λ′ ∈ suppM, there is a compatible system of isomorphisms Mλn 'Mλ′
n for all
n > m, we conclude that M = lim−→Mλn ' lim−→Mλ′
n , i.e., that suppM determines M up to
isomorphism.
(iii) Since the module Mλn defined in (i) is finite-dimensional, Mλ
n ' Vbn (λn) for
some λn ∈ h∗n. There is a unique weight µ of g such that µ ∈ 〈∆n〉R and µ|hn = λ|hn − λn. Set
242 Dimitrov and Penkov
λn := λ− µ ∈ h∗. Then M ' lim−→Mλn ' Vbn (λn) = lim−→Vbn (λn|hn ) and, according to Lemma 3,
λn belongs to an edge of the convex hull of Wn+1 · λn+1.
Theorem 7. Fix a Borel subalgebra b of g. Let {λn} ⊂ h∗ be a sequence of integral weights
of g such that λn|bn is a bn-dominant weight of gn and λn belongs to an edge of the convex
hull of Wn+1 · λn+1. Then, by the same argument as in Example 3, M := lim−→Vbn (λn|hn ) is a
well-defined irreducible weight g-module.
(i) M is finite integrable.
(ii) suppM = ∪nCλn .
(iii) M is a highest weight module with respect to some Borel subalgebra of g if
and only if there is n0 so that λn ∈W · λn0 for any n ≥ n0.
(iv) If M′ ' lim−→Vbn (λ′n|hn ), {λ′n} being a sequence of integral weights of g such
that λ′n|hn is a bn-dominant weight of gn and λ′n belongs to an edge of the convex hull of
Wn+1 · λ′n+1, then M 'M′ if and only if there is n0 so that λn = λ′n for n ≥ n0.
Proof. The proof is not difficult and is left to the reader.
Remark. The following question is quite natural and intriguing: How large is the set of
finite integrable modules which are not highest weight modules? A rigorous answer would
require a way to measure this set. One possible approach is to estimate the number of
weights λn+1 which can extend a given sequence λ1, λ2, . . . , λn. Based on Theorem 7,we can
show, using a direct combinatorial approach, that (when λ1 6= 0) any given finite sequence
λ1, λ2, . . . , λn as in Theorem 7 can be extended to an infinite such sequence. Furthermore,
if we require that suppM ⊂ 〈∆〉R (or equivalently, that λ1 ∈ 〈∆〉R) at each step, there
are only finitely many possibilities for the choice of the next weight in the sequence. A
complete combinatorial description of all possible sequences {λn} is an interesting open
problem.
Example 4. Let g = A(∞) and M = g be the adjoint module. If b is the Borel subalgebra
corresponding to the order ε1 > ε2 > . . ., thenM ' lim−→Vbn (ε1−εn+1). SinceW · (ε1−εn+1) =∆, Theorem 7(iii) implies that there exists a Borel subalgebra b such that the adjoint
module is a b-highest weight module. Furthermore, it is not difficult to verify that all
such Borel subalgebras b are precisely the Borel subalgebras which correspond to orders
on {εi} for which there exists a pair of indices i0, j0 so that εi0 > εi and εi > εj0 for all
i 6= i0, i 6= j0.
5 Pseudo highest weight modules
In the case of a finite-dimensional Lie algebra, an irreducible weight module M with
gIM = h is necessarily a highest weight module for some Borel subalgebra (see [DMP]).
Weight Modules of Direct Limit Lie Algebras 243
As we already know (Example 3), this is no longer true for the direct limit algebras we
consider. We define a pseudo highest weight module as an irreducible weight moduleM
such that gIM = h. Pseudo highest weight modules provide counterexamples also to the
obvious extension of the parabolic induction theorem to root reductive direct limit Lie
algebras. (The moduleM from Example 3 above does not provide such a counterexample
since in this case pM = g and M = U(g) ⊗U(pM) M.) Indeed, it suffices to construct M
with gFM = gIM = h such that M is not a highest weight module with respect to h ⊕ g+M
and therefore admits no surjection of the form U(g) ⊗U(pM) M′ → M for an irreducible
pM-module M′. Here is such an example.
Example 5. Let g = A(∞), B(∞), C(∞), or D(∞). Fix a Borel subalgebra b ⊂ g and let
λ ∈ h∗ be such that 〈λ, α〉 6∈ Z for all α ∈ ∆. Construct {λn} inductively by setting λ3 :=λ, λn+1 := λn + αn, where αn is a simple root of bn+1 which is not a root of bn. Since
dimVbn+1 (λn+1|hn+1 )λn|hn = 1, there is a unique (up to a constant) embedding Vbn (λn|hn )→Vbn+1 (λn+1|hn+1 ), so we can setM := lim−→Vbn (λn|hn ). Then g = h⊕g
+M⊕g
−M,where h⊕g
+M = b.
Therefore the only Borel subalgebra with respect to which M could be a highest weight
module is b. However, M has no nonzero b-highest weight vector. Indeed, such a highest
weight vector would be also a highest weight vector of Vbn (λn|hn ) for every n ≥ n0 while
Vbn (λn|hn ) do not admit a common highest weight vector for two consecutive n. Therefore,
M is not a highest weight module.
For any irreducible weight module M, we have the following natural question:
What are the integrable root directions of M? That is, for which roots α ∈ ∆ is M an
integrable gα-module? It is a remarkable fact that if ∆intM is the set of all roots α for which
M is gα-integrable, then pintM := h⊕ (⊕α∈∆int
Mgα) is a Lie subalgebra of g, and pint
M is nothing
but the subset of elements in g which act locally finitely on M. For a finite-dimensional
Lie algebra, this can be proved using Gabber’s theorem, [G]; see Corollary 2.7 in [Fe], or
Proposition 1 in [PS]. For a root direct limit Lie algebra, the statement follows immediately
from the case of a finite-dimensional Lie algebra.
Proposition 7. (i) pintM = (gFM + (gIM ∩ pint
M ))⊕ g+M.
(ii) If M is a pseudo highest weight module, then pintM = pM; in particular, pint
M is a
parabolic subalgebra of g.
Proof. (i) It is obvious that gFM ⊂ pintM and g
+M ⊂ pint
M . Furthermore, pintM ∩ g
−M = 0. Indeed,
assuming that g−M ∩ pint
M 6= 0, we would have that gα acts locally nilpotently on M for
some α ∈ ∆−M, which is impossible, as then suppM would have to be invariant with
respect to the reflection σα ∈ W. Since pintM ⊃ h, pint
M is a weight submodule of g and
pintM = pint
M ∩ ((gFM + gIM)⊕ g+M) = (gFM + (gIM ∩ pint
M ))⊕ g+M.
244 Dimitrov and Penkov
(ii) The statement follows immediately from (i) since gIM = h ⊂ gFM for a pseudo
highest weight module M.
The following proposition provides a more explicit description of pM = pintM for
highest weight modules. It is a version of the main result of [DP2] adapted to highest
weight modules with respect to arbitrary Borel subalgebras of a root reductive Lie alge-
bra g. If M = Vb(λ), we say that a root α ∈ ∆− is M-simple if and only if α = β + γ with
β, γ ∈ ∆− implies 〈λ, β〉 = 0 or 〈λ, γ〉 = 0.
Proposition 8. LetM = Vb(λ) and letΣintλ,b be the set of allM-simple roots δ ∈ ∆− for which
M is gδ-integrable. Then, for any α ∈ ∆−, M is gα-integrable (equivalently, α ∈ ∆− ∩ ∆FM)
if and only if α ∈ 〈Σintλ,b〉Z+ .
Proof. Proposition 7(ii) implies that ifα ∈ 〈Σintλ,b〉Z+ , thenM is gα-integrable. Let, conversely,
M be gα-integrable. If α is notM-simple, then α = β+γ for some β, γ ∈ ∆− with 〈λ, β〉 6= 0
and 〈λ, γ〉 6= 0. The reader will then check that |〈λ, α〉| is strictly bigger than both |〈λ, β〉| and
|〈λ, γ〉|. Choose n big enough so that α,β, γ ∈ ∆n. Applying the main theorem from [DP2] to
Vbn (λ|hn ), we obtain that Vbn (λ|hn ) is both gβ-integrable and gγ-integrable. Therefore M is
also gβ-integrable as well as gγ-integrable. To conclude that α ∈ 〈Σintλ,b〉Z+ whenever Vb(λ)
is gα-integrable, one applies now induction on |〈λ, α〉|.
Although the parabolic induction theorem does not hold for pseudo highest weight
modules, their supports can be described explicitly. In the next section, we prove a gen-
eral theorem describing the support of any irreducible weight module. An open question
about pseudo highest weight modules is whether statement (ii) of Theorem 6 extends to
any pseudo highest weight module; i.e., whether such a module is determined up to an
isomorphism by its support.
6 The support of an arbitrary irreducible weight module
Let M be an arbitrary irreducible weight module M with corresponding partition ∆ =∆FM t ∆IM t ∆+M t ∆−M. Define the small Weyl group WF of M as the Weyl group of gFM. For
λ ∈ h∗, set KλM := CλM+ 〈∆IM〉Z+ 〈∆−M〉Z+ and Kλ,nM := Cλ,nM + 〈∆IM ∩∆n〉Z+ 〈∆−M ∩∆n〉Z+ , where
CλM (resp., Cλ,nM ) is the intersection of the convex hull ofWF · λ (resp., of (WF ∩Wn) · λ) with
λ+ 〈∆〉Z.
Lemma 4. For any λ ∈ suppM and any n, there exists λn ∈ suppM such that (λ+〈∆n〉R)∩suppM = Kλn,nM .
Proof. If ∆n ⊂ ∆IM, set λn := λ. Assume now that ∆n 6⊂ ∆IM. Consider the cone K :=〈∆FM ∪∆+M〉R+ . Then (λ+ K) ∩ (suppM ∩ 〈∆n〉R) is a finite set. As in the proof of Theorem 2,
Weight Modules of Direct Limit Lie Algebras 245
suppM = ∪N suppMN for some irreducible gN-modules MN. Furthermore, there is N0
for which suppMN0 ⊃ (λ+ K) ∩ (suppM ∩ 〈∆n〉R). Let λn ∈ suppMN0 ∩ 〈∆n〉R be such that
λn+α 6∈ suppMN0 for any α ∈ ∆n∩ (∆FM∪∆+M). (Such a weight λn exists because otherwise
we would have ∆n ⊂ ∆IM.) Clearly then, (λ+ 〈∆n〉R) ∩ suppM = Kλn,nM .
Fix λ ∈ suppM. Lemma 4 enables us to construct a sequence λ1, λ2, . . . , λn, . . . such
that (λ+ 〈∆n〉R)∩ suppM = Kλn,nM . This sequence reconstructs suppM and is, in a certain
sense, assigned naturally toM. For a precise formulation, define an equivalence relation
∼M on h∗ by setting λ′ ∼M λ′′ if and only if there existsw ∈WF such thatw(λ′)−λ′′ ∈ 〈∆IM〉Z.
Let furthermore h∗M denote the set of ∼M-equivalence classes and let p : h∗ 7→ h∗M be the
natural projection.
The following theorem is our most general result describing suppM for an arbi-
trary irreducible weight g-moduleM. It is a straightforward corollary of the construction
of the sequence {λn}.Theorem 8. (i) Kλ1
M ⊂ Kλ2M ⊂ . . . and suppM = ∪nKλnM .
(ii) The sequence {p(λn)} depends on λ only, and if λ′ is another element of suppM,
there is n0 so that p(λn) = p(λ′n) for all n > n0.
(iii) If M is an irreducible weight g-module with corresponding sequence {λn},then suppM = supp M if and only if there is n0 such that p(λn) = p(λn) for n > n0.
The existence of a sequence of weights {µn} for which suppM = ∪nKµnM follows
directly from the fact that suppM = ∪n suppMn for some irreducible gn-modulesMn (see
the proof of Theorem 2). The main advantage of constructing the sequence {λn} is that
M stably determines the sequence {p(λn)}. It is an open question to describe explicitly
all possible sequences {p(λn)}, or equivalently all possible supports of irreducible weight
g-modules with a given shadow. We conclude this section with an example to Theorem 8.
Example 6. Let M = Vb(µ). Then gIM = h. Furthermore, for any fixed λ ∈ suppM, there
is N such that µ− λ ∈ 〈∆N〉R. If gFM = h, then Lemma 4 implies that λn = µ for any n ≥ N,which explains the claim of Theorem 8 in this case. If gFM 6= h, then using Proposition 8,
one checks that for n ≥ N, λn can be any element of the orbit (WF ∩Wn) · µ. This is in
agreement with Theorem 8(ii) as, in this case, the projection pM identifies all weights in
any given orbit of WF.
Appendix Borel subalgebras of g and chains of
subspaces in 〈∆〉R
In this Appendix, g is an arbitrary Lie algebra with a fixed Cartan subalgebra h such that
g admits a root decomposition (1). Here we establish a precise interrelationship between
246 Dimitrov and Penkov
Borel subalgebras b of g containing h (which we call simply Borel subalgebras), R-linear
orders on 〈∆〉R, and oriented maximal chains of vector subspaces in 〈∆〉R. Our motivation
is as follows. If g has finitely many roots (i.e., for instance, if g is finite-dimensional), then
every Borel subalgebra is defined by a (nonunique) regular hyperplane H in 〈∆〉R, i.e., by
a hyperplane H such that H ∩ ∆ = ∅. If g has infinitely many roots, this is known to be
no longer true. That is, in general, there are Borel subalgebras which do not correspond
to any regular hyperplane in 〈∆〉R. (This is the case for any Kac-Moody algebra which
is not finite-dimensional.) In [DP1], we have shown that in the case where 〈∆〉R is finite-
dimensional, every Borel subalgebra can be defined by a maximal flag of vector subspaces
in 〈∆〉R. However, when 〈∆〉R is infinite-dimensional, the situation is more complicated
and deserves a careful formulation.
We start by stating the relationship between Borel subalgebras b with b ⊃ h and
R-linear orders on 〈∆〉R.
Proposition 9. EveryR-linear order on 〈∆〉R determines a unique Borel subalgebra, and,
conversely, every Borel subalgebra is determined by an (in general not unique) R-linear
order on 〈∆〉R.
Proof. If > is an R-linear order on 〈∆〉R, then set ∆± := {α ∈ ∆ | ±α > 0}. Then ∆+ t ∆−is a triangular decomposition and thus > determines the Borel subalgebra h⊕ (⊕α∈∆+gα).
The proof of the converse statement is left to the reader (it is a standard application of
Zorn’s lemma).
In what follows,we establish a bijection betweenR-linear orders on 〈∆〉R and ori-
ented maximal chains of vector subspaces in 〈∆〉R. If V is any real vector space, a chain
of vector subspaces of V is defined as a set of subspaces F = {Fα}α∈A of V such that α 6= βimplies a proper inclusion Fα ⊂ Fβ or Fβ ⊂ Fα. The set A is then automatically ordered.
A chain F is a flag if and only if, as an ordered set, A can be identified with an (finite
or infinite) interval in Z. A chain F is maximal if and only if it is not properly contained
in any other chain. Maximal chains of vector subspaces may be somewhat counterintu-
itive as the following example shows that a vector space of countable dimension admits
noncountable maximal chains and vice versa.
Example 7. First, let V be a countable-dimensional vector space with a basis {er}r∈Q. The
chain of all subspacesVt := 〈{er | r < t}〉R for t ∈ R, V ′q := 〈{er | r ≤ q}〉R for q ∈ Q, V−∞ := 0,
and V∞ := V is a maximal chain of cardinality continuum. Second, let U := C[[x]] be the
space of formal power series in the indeterminate x. The dimension of U is continuum.
However, if F denotes the flag {Fi := xiU}i∈Z+ , then {0} ∪ F is a countable maximal chain
in U.
Weight Modules of Direct Limit Lie Algebras 247
Furthermore, it turns out that every maximal chain is determined uniquely by
a certain subchain. A maximal generalized flag2 is, by definition, a chain F which is
minimal with the following property: for every x ∈ V, there exists a pair Fα ⊂ Fβ ∈ F with
dim Fβ/Fα = 1 such that x ∈ Fβ\Fα.
Lemma 5. Every maximal chain in V contains a unique maximal generalized flag and,
conversely, every maximal generalized flag is contained in a unique maximal chain.
Proof. Let F = {Fα}α∈A be a maximal chain in V . For every nonzero x ∈ V, set Fx := ∪x 6∈FαFαand F′x := Fx⊕Rx. Then F∪{Fx, F′x} is a chain in V and hence Fx, F′x ∈ F. LetG := ∪06=x∈V{Fx, F′x}.Then G is obviously a maximal generalized flag contained in F. Noting that G consists
exactly of all pairs of subspaces from F with relative codimension one, we conclude that
G is the unique maximal generalized flag contained in F.
Let now G = {Gβ}β∈B be a maximal generalized flag in V and let < be the corre-
sponding order on B. A subset C of B is a cut of B if α ∈ C and β < α implies β ∈ C,and if α ∈ C and β 6∈ C implies β > α. The set B of all cuts of B is naturally ordered. Set
Hα := ∪β∈αGβ for α ∈ B. One checks immediately that H = {Hα}α∈B is a maximal chain in
V which contains G. On the other hand, given any maximal chain F = {Fα}α∈A in V con-
taining G, one notices that Fα = ∪Gβ⊂FαGβ. Then the map ϕ : A→ B, ϕ(α) := {β |Gβ ⊂ Fα}is an embedding of F into H and hence F = H. The lemma is proved.
An orientation of a maximal generalized flag G is a labeling of the two half-
spaces of Gβ\Gα, for every pair of indices α < β with dimGβ/Gα = 1, by mutually
opposite signs ±. An orientation of a maximal chain F is an orientation of the maximal
generalized flag it contains. A maximal chain F is oriented if and only if an orientation
of F is fixed.
Theorem 9. There is a bijection between oriented maximal chains in 〈∆〉R and R-linear
orders on 〈∆〉R.
Proof. Lemma 5 implies that it is enough to establish a bijection between oriented
maximal generalized flags in 〈∆〉R and R-linear orders on 〈∆〉R.
Let G be an oriented maximal generalized flag in 〈∆〉R. We define an R-linear
order >G on 〈∆〉R by setting x >G 0 or x <G 0 according to the sign of the half-space of
Gβ\Gα to which x belongs. Conversely, given an R-linear order > on 〈∆〉R, we build the
corresponding oriented maximal generalized flag as follows. For a nonzero vector x ∈ V,we set Vx := {z ∈ V | ±x > 0 implies ± (cz + x) > 0 for some 0 6= c ∈ R+}. Then for every
2Nonmaximal generalized flags can be defined, too. This is an interesting class of chains of subspaces whichwe intend to discuss elsewhere.
248 Dimitrov and Penkov
pair of nonzero vectors x, y ∈ V, exactly one of the following is true: Vx = Vy, Vx ⊂ Vy, or
Vx ⊃ Vy. Therefore, the set of distinct subspaces among {Vx} is a chain F. The fact that
F is a maximal generalized flag in 〈∆〉R is a direct consequence of its construction. To
complete the proof of the theorem, it remains to orient F in the obvious way.
Finally we define a decomposition of ∆,
∆ = ∆− t ∆0 t ∆+, (8)
to be parabolic if and only if Π(∆−) ∩ Π(∆+) = ∅, 0 6∈ Π(∆±), and Π(∆)\{0} = Π(∆−) t Π(∆+)
is a triangular decomposition of Π(∆)\{0} (i.e., the cone 〈Π(∆+) ∪ −Π(∆−)〉R+ contains no
vector subspace), where Π is the projection 〈∆〉R → 〈∆〉R/〈∆0〉R. If ∆0 = ∅, a parabolic
decomposition is a triangular decomposition. Given a parabolic decomposition (8), its
corresponding parabolic subalgebra is by definition p := h⊕ (⊕α∈∆0t∆+gα).
Given a linear subspace V ′ in V,we define a V ′-maximal chain of vector subspaces
in V to be the preimage in V of a maximal chain in V/V ′. A V ′-maximal chain is oriented
if and only if the corresponding maximal chain in V/V ′ is oriented. Generalizing the
corresponding construction for Borel subalgebras, we have the following theorem.
Theorem 10. Any parabolic decomposition (8) is determined by some oriented 〈∆0〉R-
maximal chain of vector subspaces in V, and, conversely, any oriented V ′-maximal chain
in 〈∆〉R, for an arbitrary subspace V ′ of 〈∆〉R, defines a unique parabolic decomposition
of ∆ with ∆0 = ∆ ∩ V ′.Proof. The proof is an exercise.
Acknowledgment
Discussions of various aspects of the subject of this paper with Yu. A. Bahturin, G. Benkart,
V. Futorny, Yu. I. Manin, O. Mathieu, G. I. Olshanskii, V. Serganova, and J. A. Wolf have been
very helpful to us. Special thanks are due to K.-H. Neeb who read an early version of the paper
and pointed out some errors. We thank also two referees whose thoughtful remarks helped make
certain points more clear.
Both authors have been supported in part by a National Science Foundation Group Infras-
tructure Grant, and the second author acknowledges support from the Louis Pasteur University in
Strasbourg where part of this work was done.
References
[BB] Yu. Bahturin and G. Benkart, “Highest weight modules for locally finite Lie algebras” in
Modular Interfaces (Riverside, CA, 1995), AMS/IP Stud. Adv. Math. 4, Amer. Math. Soc.,
Providence, 1997, 1–31.
Weight Modules of Direct Limit Lie Algebras 249
[B] N. Bourbaki, Elements de Mathematique: Groupes et algebres de Lie, Chap. 4–6, Hermann,
Paris, 1968.
[DMP] I. Dimitrov, O. Mathieu, and I. Penkov, On the structure of weight modules, to appear in
Trans. Amer. Math. Soc.
[DP1] I. Dimitrov and I. Penkov, “Partially and fully integrable modules over Lie superalgebras”
in Modular Interfaces (Riverside, CA, 1995), AMS/IP Stud. Adv. Math. 4, Amer. Math. Soc.,
Providence, 1997, 49–67.
[DP2] , Partially integrable highest weight modules, Transform. Groups 3 (1998), 241–253.
[Fe] S. Fernando, Lie algebra modules with finite dimensional weight spaces I, Trans. Amer.
Math. Soc. 322 (1990), 757–781.
[Fu] V. Futorny, The weight representations of semisimple finite dimensional Lie algebras, PhD
thesis, Kiev University, 1987.
[G] O. Gabber, The integrability of the characteristic variety, Amer. J. Math. 103 (1981), 445–
468.
[Ha] A. Habib, Direct limits of Zuckerman derived functor modules, preprint, 1998.
[Hu] J. E. Humphreys, Introduction to Lie Algebras and Representation Theory, Grad. Texts in
Math. 9, Springer-Verlag, New York, 1972.
[M] O. Mathieu, Classification of irreducible weight modules, preprint, 1997.
[Na] L. Natarajan, Unitary highest weight modules of inductive limit Lie algebras and groups,
J. Algebra 167 (1994), 9–28.
[NRW] L. Natarajan, E. Rodriguez-Carrington, and J. A. Wolf, The Bott-Borel-Weil theorem for
direct limit Lie groups, to appear in Trans. Amer. Math. Soc.
[Ne] K.-H. Neeb, Holomorphic highest weight representations of infinite dimensional complex
classical groups, J. Reine Angew. Math. 497 (1998), 171–222.
[O1] G. I. Olshanskii, Unitary representations of the group SO0(∞,∞) as limits of unitary rep-
resentations of the groups SO0(n,∞) as n→∞, Funct. Anal. Appl. 20 (1988), 292–301.
[O2] , “Representations of infinite-dimensional classical groups, limits of enveloping
algebras, and Yangians” in Topics in Representation Theory, Adv. Soviet Math. 2, Amer.
Math. Soc., Providence, 1991, 1–66.
[P] T. D. Palev, Highest weight irreducible unitary representations of Lie algebras of infinite
matrices, I: The algebra gl(∞), J. Math. Phys. 31 (1990), 579–586.
[PS] I. Penkov,V. Serganova,The support of an irreducible Lie algebra representation, J. Algebra
209 (1998), 129–142.
[Zh] D. Zhelobenko, Representations of reductive Lie algebras (in Russian), Nauka, Moscow,
1994.
Dimitrov: Department of Mathematics,University of California,Los Angeles,California 90095,USA;
Penkov: Department of Mathematics, University of California, Riverside, California 92521, USA;